How many reels were analysed, and how?

3,527 reels with full metrics, pulled across 123 hashtags and 21 niches so the sample covers the whole performance range rather than only content that already succeeded. Of those, 1,614 were blind-coded for hook, emotion and format by raters who never saw the view counts. Duration and caption length were computed on all 3,527.

What is the difference between a reliable lever and a lottery lever?

A reliable lever raises the median reel: almost every post improves, with low variance. A lottery lever has a very high ceiling but a poor median, so most attempts go nowhere and a rare one explodes. Curiosity and surprise emotion, value hooks and short duration are reliable. Awe emotion (34.7x variance), cinematic format and the travel niche (40.1x) are lotteries. Most advice fails because it mixes the two.

What should a hotel or restaurant post instead?

Four reliable formats: a tutorial showing how the kitchen builds a dish (32,294 median), a useful list of nearby places (34,842, the best median of any hook), a before-and-after such as a room being turned over (33,341), and a disbelief hook like 'this costs less than you think' (34,443). All four promise the viewer something in the first two seconds. Add on-screen text, which lifts the median from 17,828 to 31,588, and keep it under 15 seconds.

How reliable are these findings?

A model trained on the data predicts breakout at AUC 0.69, meaning content explains roughly a third of the outcome and the rest is account size, distribution and luck. Stacking the levers improves the hit rate by around 31% in relative terms. It is not a guarantee. Posting day and hour are confounded with post age and are not reported as reliable.